DeepSeek’s reported plan for an in-house AI chip adds pressure on Nvidia, sending NVDA lower

Nvidia’s stock fell as investors digested a new report that DeepSeek, the Chinese AI lab and model developer, is working on its own in-house artificial-intelligence chip. The move, described in market coverage citing the chip effort, added to concerns that some customers may increasingly prefer custom silicon rather than buying Nvidia’s chips for training and running large-scale AI workloads.

The report said DeepSeek aims to develop a bespoke chip designed to support its AI systems, and it framed the effort as a way to reduce reliance on external suppliers. That prospect matters for Nvidia because Nvidia’s data-center revenue is tightly linked to how many AI builders choose its GPUs and related platforms for model training and deployment.

Market participants also appeared to connect the news to broader competitive pressure in AI compute. Custom chips can potentially lower costs for a specific provider and can be optimized for that provider’s model architectures and software stack. Even when such chips do not fully replace third-party accelerators, partial substitution can change procurement patterns over time.

Nvidia stock was reportedly headed toward its lowest level since April following the report. The reaction underscores how sensitive Nvidia’s valuation is to incremental changes in perceived demand for its AI accelerators, particularly when the market interprets them as a announcement of accelerating in-house chip development among major AI labs.

Beyond the immediate story, the episode highlights a recurring tension in the AI chip market. Nvidia has benefited from a dominant position in accelerating training and inference with its GPU-based systems, but the same scale of demand that attracted custom hardware also encourages large-scale AI companies to experiment with alternatives.

Still, the details investors will want are not fully clear from the market report. It did not provide, in the coverage available here, specifics on the chip’s timeline, performance relative to Nvidia’s latest accelerators, manufacturing approach, or whether DeepSeek intends to scale production for internal use only or for broader sale.

For now, the key question is how quickly any custom-chip program translates into reduced purchasing of Nvidia hardware and whether it affects both training and inference workloads. Nvidia did not disclose any response in the materials reviewed for this report, and the report itself did not establish those operational specifics.

What to watch next is whether Nvidia and major AI customers offer more concrete information about their procurement plans and hardware roadmaps, including any public benchmarks, deployment milestones, or changes in capacity allocation between Nvidia chips and custom silicon.